Resource allocation from base stations to mobile users in realistic MIMO-OFDMA systems such as the 3GPP Long Term Evolution (LTE) downlink is based on limited and quantized channel feedback over a fine-granular resource grid of multiple dimensions. This allows for opportunistic scheduling but impedes application of enhanced cross-layer strategies due to the discrete and combinatorial problem space. Integer optimization for this allocation problem is strongly complex and prohibits use of efficient algorithms. Provided solutions in practice are given by sub-optimal greedy heuristics. In this paper, we apply two-sided stable matchings for adaptive multi-user scheduling. Our framework gives Pareto-efficient allocations and yields a tunable tradeoff between system throughput and user fairness. We form stable pairings of system resources and users based on queue-and channel-aware lists of preferred matches. The derived concept aims to find a stable matching state under presence of non-strict preference relations if such exist or redefines the allocation problem to a solvable strict problem instance. A performance evaluation for scheduling is done by system level simulations for high traffic loads in a realistically modeled LTE deployment.